Install

Drop it in. Reference it. Ship.

Deliberate is plain markdown. No build step, no dependencies, no lock-in. Works with any agent that can read a file.

01

One command. Installs all eight skills into your project for Cursor, Claude Code, Gemini CLI, Codex CLI, and 40+ other agents.

npx skills add angad729/deliberate

Works with Cursor, Claude Code, Codex, Gemini, Windsurf, Cline, Continue, Goose, Kilo, Roo, and more. Powered by vercel-labs/skills.

02

Or copy the files manually

Prefer to see and control what lands in your repo? Clone and copy.

git clone https://github.com/angad729/deliberate.git /tmp/deliberate
cp -r /tmp/deliberate/skills ./

Claude Code

Create or update CLAUDE.md at your project root:

# CLAUDE.md

Follow the guidelines in @skills/deliberate/SKILL.md for all code changes.
When working on features against a PRD, also follow @skills/spec/SKILL.md.
When debugging, also follow @skills/debug/SKILL.md.
When reviewing PRs, also follow @skills/review/SKILL.md.
When writing tests, also follow @skills/test/SKILL.md.
When crossing component boundaries, also follow @skills/architect/SKILL.md.
When planning a migration, also follow @skills/migrate/SKILL.md.
When responding to an incident, also follow @skills/incident/SKILL.md.

Cursor

Drop the skill folders into .cursor/rules/, or reference from .cursorrules.

Codex / Gemini / other agents

Include the files in your project context or system prompt. Plain markdown with frontmatter works everywhere.

Paste-in fallback

No config? Paste the contents of skills/deliberate/SKILL.md into your agent's system prompt or custom instructions. Done.

03

Load skills situationally, not all at once

deliberate.md is the default load. Pull in a specialized skill only when the task calls for it:

Loading every skill for every task defeats the point - each skill is a lens, and too many lenses at once blurs the focus.


When not to use it
  • Trivial one-liners. Deliberate biases toward caution over speed. For throwaway scripts or obvious edits, the overhead isn't worth it.
  • Rapid prototyping where correctness doesn't matter yet. Plan-first gets in the way of exploration.
  • You disagree with a principle. Fork it. Delete sections. Rewrite for your taste. Starting point, not scripture.